A real-time crop lodging recognition method for combine harvesters based on machine vision and modified DeepLab V3+

作者：Cong Yao, Dawei Lv, Hua Li, J. L. Fu, Chao Li, Xiaojun Gao, Daolong Hong · 发表于：Smart Agricultural Technology · 年份：2025 · DOI：10.1016/j.atech.2025.100926 · 被引用次数：12 · 研究领域：Smart Agriculture and AI、Industrial Vision Systems and Defect Detection、Remote Sensing in Agriculture

To minimize the losses during the harvest of lodged crops, an effective recognition method of crop lodging is particularly important for combine harvesters . This study presents a method for real-time monitoring of crop lodging using machine vision and semantic segmentation, offering an alternative to traditional manual inspection techniques. Firstly, the DeepLab V3+ model was applied to the recognition of lodging areas, with modifications made to meet the harvesting requirements. Xception served as the backbone network during training to improve accuracy, while MobileNet V2 was adopted during deployment to balance accuracy with computational efficiency. Next, a RealSense depth camera was installed on the combine harvester cabin to enable image data collection and crop height recognition. Finally, a ROS (Robot Operation System) framework was designed and implemented on a Jetson Nano board to integrate lodging area recognition with crop height measurement. Experimental results demonstrated that the proposed method achieved a recognition pixel accuracy (PA) of 93.83% and a mean intersection over union (mIoU) of 85.3% for lodging areas, and 93.77% PA and 83.97% mIoU for non-lodging areas. Crop height recognition errors below 4%, meeting the standards required for real-time recognition. In general, this solution proved to be cost-effective, easy to deploy, and highly accurate, making it well suited for practical application in harvesters and offering valuable insight to reduce ha...